Posts

How Does Eco-Coaching Help to Save Energy? Assessing a Recommendation System for Energy-Efficient Thermostat Scheduling

AbstractEco coaching: Field deployment for exploring a design approach; for specific actions that would reduce wasted energy. ThermoCoach: Eco-coaching for thermostat scheduling.Senses and models occupancy patterns in a home & provides suggestions for configuring thermostatEco coaching accomplishes four thingsUsers can implement effective thermostat scheduleIt supported user agency in negotiating energy savings and goalsFacilitated learning different scheduling strategies and weighing different optionsChallenged user beliefs about how well they were doing. IntroTechnology in helping energy savingEco-feedbackGive info about energy usageSome actions are too complex or time consuming AutomationSave energy on users behalfNot work as well (why?) Eco-coachingOne step more than eco-feedback; provide feedback based on past data but also identify waste and recommend actions to prevent energy waste in the future. One step less than automationMixed initiativeThermocoachdeployment of eco coac…

Reading "High-dimensional Time Series Clustering via Cross-Predictability"

AbstractTime series clustering : similarity metric is key"Cross-predictability" -> new metric proposed by this paper; degree to which a future value in each time series is predicted by past values of the others. Challenging in high-dimensional regime -> Sparsity assumption (only time-series in the same cluster have significant cross-predictability)This paper = first practical high-dimensional time-series clustering with a provable guaranteeIntroIn IoT, # sensors >> # readings in the time series (d >> T); under-constrained (d variables and T equations) Time series clustering metrics -> either not provable or considers non-high dimensional"cross-predictability" : how much a future value is predicted by past values of the others Captures causal relationships between time seriesDeal with under-constrained problem with sparsity assumption (only time series in the same cluster have significant cross-predictability with each other)Regularized Dantzig …

Reading "Transferring Decomposed Tensors for Scalable Energy Breakdown across Regions"

AbstractEnergy breakdown solutions require (1) instrumentation (2) model training for each geographical region This paperRegion independent energy breakdown model via statistical transfer learningIntuitionHeterogeneity in homes and weathers across different regions most significantly impacts the energy consumption across regions Factor out heterogeneity to learn homogeneous energy breakdown components for each individual appliance.Transfer model learned in one region to another regionIntroductionEnergy breakdown benefits -> energy saving, load forecasting, improving design, policy makingHow to energy breakdownMetering hardware appliance-levelNILM -> Hardware home-level, source separationNNMF -> Benefit: Similar instrumentidea: Similar homes have similar per-appliance energy breakdown. So, estimate breakdown of a home based on a similar home which already has breakdown available. Problem: Has IID assumption between test / training and requires testing home to be similar to tra…

Reading "Matrix Factorisation for Scalable Energy Breakdown"

AbstractEnergy breakdown is a valuable information for reducing energy consumption Existing approaches = expensive, requires hardwareThis paperFeature based matrix factorization for handling repeating structure in energy data due to common design and construction patterns Evaluate on 516 homes from public data; better than 5 baselines Deployment of system as a live web application to provide energy breakdown IntroImportance of energy breakdownEnergy breakdown is expensive Domain: collaborative filtering through feature-based matrix factorization for energy breakdownDataport data set with 516 homes Related workScalability of energy breakdown with (1) metering hardware (2) energy disaggregation techniques ; they all require hardware Direct load metering - requires power meter DisaggregationInfer the power of energy of individual loads based on aggregate power measurements Under-constrained problem.Requires source separation techniques Requires meter per building gathering data at freque…

(Coursera) (Week 1) Discrete Optimization

Week 1NP completePolynomial time solution checkingSolving 1 NP complete problem solves all other NP complete problemsReduction of 1 problem to anotherIn practice, we solve small problems, "push the exponential", so we can solve the practical problems Examples Kidney Exchanges Pair of D / R (Doner, Receiver), that are not compatible Week 2ModelingDefine precise description of the problem that everyone can agree uponInputGoal (What to optimize) Constraints StepsChoose decision variablesSomething that captures the real decisions you're interested in Express the problem constraints in terms of the variablesSpecify the solutions to the problemExpress the objective function Specify quality of each solution KnapsackDecision ProblemXi = 1 : Item is selectedXi = 0 : Item is not selected Constraint\sum_{i} w_i x_i \leq KObjective Function \sum_{i} v_i x_iMaximize objective subject to constraint What's left? Find values for the decision variablesNumber of solutions : 2^|I| ; t…

(Coursera) (Week 6) Spatial Data Science and Applications

6.1 Desktop GIS
Limitations Simple spatial analysis of demand & supply gives an insight of timber land investment Assumption for no activities over boundaries
6.2 Server GIS  Use caseIntegrated municipal spatial databasesWhen is Server GIS appropriate?Multiple actors with different desired rolesData is managed at the central DBMS, with DBMS features WorkflowSet SDBMSPostgresSQL, PostGISCreate UsersSet PrivilegeUpload datasetView & Analyze data using QGIS https://learnosm.org/it/osm-data/setting-up-postgresql/6.3 Spatial Data Analytics I Spatial DependencyMOHW (Ministry of Health and Welfare) wants to check any spatial relationships between districts and disease prevalence rate. Influential Variable DetectionMOHW also wnat to see if there's any regional factors that influence disease prevalence rateSolutionsFirst stage: Spatial Autocorrelation Analysis Conducted with respect to disease prevalence rate of adminstrative districtFinds the list of diseases with spatial autocorrelat…

(Coursera) (Week 5) Spatial Data Science and Applications

https://www.coursera.org/learn/spatial-data-science

Week 5 seemed most interesting since it covers some of math and science not 

5.1 Introduction


What is spatial data analyticsScience of processing spatial data with the goal of discovering useful informationCategorization (based on input data and outcome) of Spatial Data Analysis Measurement & Basic GeoProcessing Proximity and Accessibility AnalysisSpatial AutocorrelationSpatial Interpolation Spatial Categorization Clusteirng & HotspotFactor AnalysisTerrain AnalysisNetwork AnalysisSpatio-temporal data mining Example: Trajectory analysis 
5.1 Proximity and Accessibility
Goal : Determine the distance relationship between selected feature and other features Demand vs. SupplyDemand: What's the closest store from my placeSupply : What's the area that the place can reach to Thiessen Polygons / Voronoi diagram Partitioning of plane with input points into polygonsEach polygon contains exactly 1 point Boundaries define the area that…